Novel Semiconductor Chips for Neuromorphic Computing

The human brain with its neuronal networks, consisting of millions of neurons, served as model for researchers, as they set about depicting them in the circuits of neuromorphic chips. Neuronal networks are applied as algorithms for integrated circuits here, for imitating neurobiological architectures. The data is calculated in parallel distributed memory units instead of Central Processing Units (CPUs) used so far. This makes neuromorphic chips significantly faster and  more efficient than the currently used processors.

Neuromorphic Computing
© Fotolia
Neuromorphic Computing

The Fraunhofer EMFT researchers bring in their competences in the area of micro- and nanotechnologies for developing novel neuromorphic systems for semiconductor chips. For example in the EU-funded project NEURONN, Fraunhofer EMFT is developing a new computer architecture inspired by neuronal networks, together with six european partners. Here novel 2D materials are deployed for implementing memristive components - consisting of memory and resistor - which are then used als synapses of the system. In the TEMPO project, also funded by the EU, the Fraunhofer EMFT circuit design team is utilizing new integrated memory technologies in innovative concepts for implementation of analog and digital neuromorphic circuits. 

In neuromorphic computing, the laborious data transfer between processor and memory can be omitted, which minimizes latency times as well as power consumption for complex calculation and transfer processes. This makes the technology interesting especially for AI-applications, since it can help to solve one of the most urgent challenges in digitalization: the huge energy consumption of computer centers. 

Neuromorphic chips are predestined for applications benefitting from increased energy-efficiency and short latency times, e.g. in medical technology for analysis and processing of biosignals like EKG or EEG, or for speech analysis and hearing aids. The technology can also facilitate signal processing and AI-supported data analysis for sensor applications by significantly increasing the energy-efficiency in mobile and portable sensor systems, e.g. for autonomous driving, condition monitoring, in space applications, or in the so-called "electronic nose" for detection of gases and smells.

You could also be interested in:


2D Materials


Memristive Components


IC Design and Layout

Project: Power-saving Chips for Neuromorphic Computing

Project: Energy-efficient Neuromorphic Computing